Mounted Single Particle Characterization for 3D Mineralogical Analysis—MSPaCMAn

Round 1

Reviewer 1 Report

# General

+ The authors present a methodology for quantifying mineral phases in 3D using X-ray micro computed tomography (XMT).

+ This task has been addressed by many authors in the past decade, as the authors well summarised in their Introduction. However, they seem to have missed out one approach in their review, namely dual-energy X-ray micro computed tomography:

[1] Lin, C. L., Hsieh, C. H., Tserendagva, T. A., & Miller, J. D. (2013). Dual energy rapid scan radiography for geometallurgy evaluation and isolation of trace mineral particles. Minerals Engineering, 40, 30–37. https://doi.org/10.1016/j.mineng.2012.09.001

[2] Tsuchiyama, A., Nakano, T., Uesugi, K., Uesugi, M., Takeuchi, A., Suzuki, Y., Noguchi, R., Matsumoto, T., Matsuno, J., Nagano, T., Imai, Y., Nakamura, T., Ogami, T., Noguchi, T., Abe, M., Yada, T., & Fujimura, A. (2013). Analytical dual-energy microtomography: A new method for obtaining three-dimensional mineral phase images and its application to Hayabusa samples. Geochimica et Cosmochimica Acta, 116, 5–16. https://doi.org/10.1016/j.gca.2012.11.036

+ The methodology is tested for three different mineral samples, that provides sufficient variety of minerals of interest and challenges known to occur. However, in my opinion the claim of “unprecedented” ability to differentiate mineral phases seems overstated and not supported by what is on the article. Is the segmentation by MSPaCMAn unprecedented because of the number of mineral phases or because it differentiates pyrite from chalcopyrite? The former is not as clear-cut as mineralogy and texture of the sample influences the number of phases that can be distinguished, as it is shown in the paper by the authors. The latter case has been done before with dual-energy or by correlating XMT and MLA. What I can indeed concede to the authors is the novelty of doing the analysis in a particle-by-particle basis, and thus include additional criteria. However, as mentioned before, it is not clear to the reader that the result achieved is “unprecedented”

+ I’d suggest using the acronym XMT instead of CT when linking it to X-ray micro-tomography. CT is mostly used to describe the medical scanning system, that obviously shares the technology and methods to XMT but differentiates from it in the voxel size and application.

+ The methods are described in a very detailed way; I thank the authors for this effort. However, in sections 2.4.1, 2.4.2 and 2.4.3 the number of projections and scanning time should also be included. The size of the marker used for the watershed algorithm would also be a nice addition.

+ Can the authors comment on how would they expect their method to perform at coarse particle sizes, where not much liberation is expected. Less liberation mean that the particle’s histogram will become similar to the bulk histogram (extreme case of just one big particle being scanned). Some grade engineering or early waste rejection applications work at particle sizes of -10+1mm where only a few particles can be included in the field of view if one doesn’t want to compromise in voxel size. The examples provided by the authors seem to be on the fairly-liberated side of the spectrum, helping the case for doing a particle-based analysis. A comment on this is welcomed.

 

# Specific

+ The name of the methodology is wrongly spelled in the abstract, please correct.

+ in line 17, I’d suggest changing the phrase “, a reduced detection limit of phases” into an affirmative phrase, like “an increased number of detected phases”. It sounds to me that with MSPaCMAn one achieves fewer phases as the limit is lower

+ in line 27: Why “affordable”. Isn’t the demand for resources in general expected to increase? Not just for affordable ones.

+ in line 28: change to rare-earth elements

+ suggest changing to “steady decrease in ore grades”

+ In line 32 there is a logical jump between those sentences. It is not clear why a 3D particle-based characterisation will improve minerals processing technology’s efficiency. The link is somehow outlined in the paragraph immediately after when the authors highlight the importance of variability and ore texture (without mentioning these explicitly, see comment below).

+ In paragraph 2 (lines 35 to 52) the authors highlight the importance of analysing properties at the particle-scale rather than at the bulk-scale. However, the authors missed the opportunity to mention two key aspects, ore texture and its variability at different scales.

+ Line 51: change “third dimension” to “a three-dimensional analysis”—I don’t think the authors mean associating particle properties to the 3^rd dimension (width)

+ Line 66: This workflow includes…

+ Line 71: See comment above about “unprecedented”

+ Line 91: What do the authors mean by “circularity”? I guess the authors mean “size”. Or do they mean “circularity” as a shape descriptor (how circular a shape is)? The former makes sense, as the source is far from a theoretical point (no size) but might be actually quite circular

+ Line 108: the use of the adjective “inapplicable” might be a bit harsh. Current methods can be applied, but they might be limited to certain particle sizes. Also, bear in mind that particle size in mineral processing varies over from a few centimetres (SAG primary crusher) down to a few hundred microns (cyclone overflow, regrind, etc). Are the authors claiming then that current methods cannot be applied to any of those particle sizes?

+ Line 121. Change to Therefore. Thereafter has a temporal logic that does not seem to apply the logic flow of the sentences

+ Line 137. Ditto

+ Line 161: Do the authors mean “smaller”? One aims for a voxel size smaller than the object of interest, no?

+ Line 163: The filter’s material can also be changed if required.

+ Line 170: Maybe worth referencing a couple of papers. I grant you that Gy’s law might be controversial, though.

• Evans, C. L., & Napier-Munn, T. J. (2013). Estimating error in measurements of mineral grain size distribution. Minerals Engineering, 52, 198–203. https://doi.org/10.1016/j.mineng.2013.09.005
• Gy, P., 1979. Sampling of Particulate Materials. Theory and Practice. Elsevier, Amsterdm. 431 pp

+ Line 176. Breaking down

+ Line 180. Seek a way to differentiate the physical scanning filters and the image filters

+ Line 193: Change to “quency in a particle, namely its pixel intensity histogram.

+ Line 215: fewer peaks

+ line 235: Ref [14] also addresses this issue. Include alongside 22 and 25

+ Line 260: insert “target” between “the” and “material”, and remove “to be analysed”

+ Line 332: Maybe mention how this translates in the total number of particles per scan. As this is the factor to be considered when deciding how many samples to prepare for a specific ore

+ Line 336: ParVox is not discussed in the surrounding text of Table 1, thus not introduced before being referenced pages later in the text.

+ Line 439: The flat shape of the particles broadens the peak of muscovite, why? PVE? Please explain

+ Figure 6: Supplementary material explains how the Eeff was obtained, and the paper explains how to obtain the attenuation plot from the NIST database. However, it is not clear to me how the superimposition of the grey-scale histogram was obtained. There is a relation between grey-scale values and material’s attenuation, however this relation is seldom known explicitly. How did the authors find this relation so the histogram could be plotted on top of the attenuation plot?

+ Line 475: Maybe add a reference to Table 1. First time ParVox is mentioned in the text and Table 1 is a few pages upstream

+ Line 482: please check colours. Feldspar appears yellow and mica looks light blue to my eyes

+ Line 490: Is 28 particles statistically significant? Also, is 28 particles for each phase or in total for all three phases?

+ Line 622: criteria, not criterial

+ Line 626: what’s shown in Fig 9 is not a line, but a region.

+ Line 652: again “unprecedented” I need to be more convinced

Author Response

We appreciate the time dedicated to the very productive review of our article. We have largely followed your suggestions, with some remarks about the comparison of our method to dual energy

# General

+ The authors present a methodology for quantifying mineral phases in 3D using X-ray micro computed tomography (XMT).

Reply: In fact, our paper focus on improving phase classification, a necessary pre-step towards accurate quantification. Not so many studies specifically address the classification issue as suggested in the next point.

+ This task has been addressed by many authors in the past decade, as the authors well summarised in their Introduction. However, they seem to have missed out one approach in their review, namely dual-energy X-ray micro computed tomography:

[1] Lin, C. L., Hsieh, C. H., Tserendagva, T. A., & Miller, J. D. (2013). Dual energy rapid scan radiography for geometallurgy evaluation and isolation of trace mineral particles. Minerals Engineering, 40, 30–37. https://doi.org/10.1016/j.mineng.2012.09.001

[2] Tsuchiyama, A., Nakano, T., Uesugi, K., Uesugi, M., Takeuchi, A., Suzuki, Y., Noguchi, R., Matsumoto, T., Matsuno, J., Nagano, T., Imai, Y., Nakamura, T., Ogami, T., Noguchi, T., Abe, M., Yada, T., & Fujimura, A. (2013). Analytical dual-energy microtomography: A new method for obtaining three-dimensional mineral phase images and its application to Hayabusa samples. Geochimica et Cosmochimica Acta, 116, 5–16. https://doi.org/10.1016/j.gca.2012.11.036

Reply: We would like to note that dual energy can also be used coupled to our method and we agree that is worth referencing. However, we do not think those references suggested by the reviewer are the most appropriate, reference [1] is about a radiograph technique (2D), and reference [2] is about synchrotron based X-rays that has very different types of imaging artefacts. Both references are difficult to relate to our work. Instead, we added this reference as we believe it is an example of dual energy laboratory CT that can be more easily related to our work and the context of our introduction: DOI:10.3390/s21072455.

+ The methodology is tested for three different mineral samples, that provides sufficient variety of minerals of interest and challenges known to occur. However, in my opinion the claim of “unprecedented” ability to differentiate mineral phases seems overstated and not supported by what is on the article. Is the segmentation by MSPaCMAn unprecedented because of the number of mineral phases or because it differentiates pyrite from chalcopyrite? The former is not as clear-cut as mineralogy and texture of the sample influences the number of phases that can be distinguished, as it is shown in the paper by the authors. The latter case has been done before with dual-energy or by correlating XMT and MLA. What I can indeed concede to the authors is the novelty of doing the analysis in a particle-by-particle basis, and thus include additional criteria. However, as mentioned before, it is not clear to the reader that the result achieved is “unprecedented”

Reply: Ok, we removed the word “unprecedented”. The reviewer is right, we cannot claim unprecedented resolution because we do not show quantitative proof. We do, however, show an improved classification ability (qualitative) that has a more universal applicability than available classification techniques.

Please note that despite the different methods found in the literature, to our knowledge. none shows a general improvement in the ability to classify phases but rather punctual improvements specific of a case study. For example, dual energy, works well to improve the contrast between two specific phases for which the scanning energies must be handpicked. Even when increasing the contrast between two phases, the same conditions may also compromise the contrast between other phases, thus it is not a solution towards a general 3D characterization method. Dual energy works well in medicine because there is a limited amount of materials that can be found inside humans (e.g. bone/meat), thus it is easy to standardize the scanning parameters.

+ I’d suggest using the acronym XMT instead of CT when linking it to X-ray micro-tomography. CT is mostly used to describe the medical scanning system, that obviously shares the technology and methods to XMT but differentiates from it in the voxel size and application.

Reply: We wish there was a consensus in the literature about a unique acronym for the technique. But there isn’t, therefore, we rather stick to our group’s most used acronym (CT) in previous papers, which is also shared by other imaging groups. Additionally, we follow the journal specification for defining acronyms in the text.

+ The methods are described in a very detailed way; I thank the authors for this effort. However, in sections 2.4.1, 2.4.2 and 2.4.3 the number of projections and scanning time should also be included. The size of the marker used for the watershed algorithm would also be a nice addition.

Reply: The number of projections and the scanning time were add. The H-maxima value used for the watershed was 2 as defined in reference [35].

+ Can the authors comment on how would they expect their method to perform at coarse particle sizes, where not much liberation is expected. Less liberation mean that the particle’s histogram will become similar to the bulk histogram (extreme case of just one big particle being scanned). Some grade engineering or early waste rejection applications work at particle sizes of -10+1mm where only a few particles can be included in the field of view if one doesn’t want to compromise in voxel size. The examples provided by the authors seem to be on the fairly-liberated side of the spectrum, helping the case for doing a particle-based analysis. A comment on this is welcomed.

Reply: We added additional discussion in section 4.4. In principle, the larger the particles are, the lower the capacity to homogenize the sample, unless its diameter is increased, which would compromise the voxel size. Nevertheless, even for non-liberated particles a simplification of the particle histogram relative to the all-particles histogram would still be expected. Similarly, the larger the particles are, the more complex the histogram is expected.

 # Specific

+ The name of the methodology is wrongly spelled in the abstract, please correct.

Reply: Corrected

+ in line 17, I’d suggest changing the phrase “, a reduced detection limit of phases” into an affirmative phrase, like “an increased number of detected phases”. It sounds to me that with MSPaCMAn one achieves fewer phases as the limit is lower

Reply: Done

+ in line 27: Why “affordable”. Isn’t the demand for resources in general expected to increase? Not just for affordable ones.

Reply: Ok, true. But in practice only the affordable ones are used J

+ in line 28: change to rare-earth elements

Reply: Done

+ suggest changing to “steady decrease in ore grades”

Reply: Done

+ In line 32 there is a logical jump between those sentences. It is not clear why a 3D particle-based characterisation will improve minerals processing technology’s efficiency. The link is somehow outlined in the paragraph immediately after when the authors highlight the importance of variability and ore texture (without mentioning these explicitly, see comment below).

+ In paragraph 2 (lines 35 to 52) the authors highlight the importance of analysing properties at the particle-scale rather than at the bulk-scale. However, the authors missed the opportunity to mention two key aspects, ore texture and its variability at different scales.

Reply: We now mention texture and its variability and decreased the logical jump between sentences.

+ Line 51: change “third dimension” to “a three-dimensional analysis”—I don’t think the authors mean associating particle properties to the 3^rd dimension (width)

Reply: Done

+ Line 66: This workflow includes…

Reply: Done

+ Line 71: See comment above about “unprecedented”

Reply: Changed the word to improved.

+ Line 91: What do the authors mean by “circularity”? I guess the authors mean “size”. Or do they mean “circularity” as a shape descriptor (how circular a shape is)? The former makes sense, as the source is far from a theoretical point (no size) but might be actually quite circular

Reply: The reviewer is right, size and shape contribute. In practice the source is typically an ellipse with heterogeneous intensity due to the angle between the electron gun and the X-ray target.

+ Line 108: the use of the adjective “inapplicable” might be a bit harsh. Current methods can be applied, but they might be limited to certain particle sizes. Also, bear in mind that particle size in mineral processing varies over from a few centimetres (SAG primary crusher) down to a few hundred microns (cyclone overflow, regrind, etc). Are the authors claiming then that current methods cannot be applied to any of those particle sizes?

Reply: “inapplicable” was indeed too harsh and we have moderated the sentence.

+ Line 121. Change to Therefore. Thereafter has a temporal logic that does not seem to apply the logic flow of the sentences

+ Line 137. Ditto

Reply: Done, done

+ Line 161: Do the authors mean “smaller”? One aims for a voxel size smaller than the object of interest, no?

Reply: Yes, Corrected

+ Line 163: The filter’s material can also be changed if required.

Reply: Yes, info was add

+ Line 170: Maybe worth referencing a couple of papers. I grant you that Gy’s law might be controversial, though.

• Evans, C. L., & Napier-Munn, T. J. (2013). Estimating error in measurements of mineral grain size distribution. Minerals Engineering, 52, 198–203. https://doi.org/10.1016/j.mineng.2013.09.005
• Gy, P., 1979. Sampling of Particulate Materials. Theory and Practice. Elsevier, Amsterdm. 431 pp

Reply: Evans13 was add to the references. Also Blannin21 fits here well.

+ Line 176. Breaking down

Reply: Done

+ Line 180. Seek a way to differentiate the physical scanning filters and the image filters

Reply: Image filter is now used throughout the paper.

+ Line 193: Change to “quency in a particle, namely its pixel intensity histogram.

Reply: we changed it to voxel intensity histogram

+ Line 215: fewer peaks

Reply: Corrected

+ line 235: Ref [14] also addresses this issue. Include alongside 22 and 25

Reply: Indeed, reference added

+ Line 260: insert “target” between “the” and “material”, and remove “to be analysed”

Reply: Done

+ Line 332: Maybe mention how this translates in the total number of particles per scan. As this is the factor to be considered when deciding how many samples to prepare for a specific ore

Reply: We disagree with using particles per scan because the volume of a scan may change with the scan settings and is not comparable for different scanners where the detectors have different number of pixels and resolution. We keep particles/volume as a more comparable unit.

+ Line 336: ParVox is not discussed in the surrounding text of Table 1, thus not introduced before being referenced pages later in the text.

Reply: We added two sentences to relate the ParVox to figures 2 and 3 in section 3.1.

+ Line 439: The flat shape of the particles broadens the peak of muscovite, why? PVE? Please explain

Reply: Yes, mainly PVE, this information was added.

+ Figure 6: Supplementary material explains how the Eeff was obtained, and the paper explains how to obtain the attenuation plot from the NIST database. However, it is not clear to me how the superimposition of the grey-scale histogram was obtained. There is a relation between grey-scale values and material’s attenuation, however this relation is seldom known explicitly. How did the authors find this relation so the histogram could be plotted on top of the attenuation plot?

Reply: The relation between grey-scale and attenuation coefficient should be known and is given by the reconstruction software. We are aware that some reconstruction software are unknown black boxes. However, in our software the maximum CT value (input of the reconstruction that corresponds to calculated attenuation coefficients of voxels) corresponds to the grey-value of 65535 (maximum of a 16-bit image). A direct conversion can be done (assuming no image artefacts, or the artefacts are equally affecting all phases in all particles – thus affecting equally the relative grey-scale).

+ Line 475: Maybe add a reference to Table 1. First time ParVox is mentioned in the text and Table 1 is a few pages upstream

Reply: Several references to table 1 were add in the different sections.

+ Line 482: please check colours. Feldspar appears yellow and mica looks light blue to my eyes

Reply: Yes, this was corrected.

+ Line 490: Is 28 particles statistically significant? Also, is 28 particles for each phase or in total for all three phases?

Reply: Changed to “10 mica, 9 quartz and 9 feldspar particles”. These are the number of particles of those phases available in the analysed MLA cross-section. Here, we intend to demonstrate the possibilities of the technique to classify few particles and further statistical/quantitative studies will follow.

+ Line 622: criteria, not criterial

Reply: Corrected

+ Line 626: what’s shown in Fig 9 is not a line, but a region.

Reply: Corrected

+ Line 652: again “unprecedented” I need to be more convinced

Reply: Changed to high phase resolution

Reviewer 2 Report

This has the potential to be a very interesting piece of work, but at the moment it feels like there are missing parts to the paper.  The abstract and introduction suggest a methods paper (in particular sample preparation and particle characterisation) but there is limited detail on the rationale for the preparation methods chosen and validation of the particle characterisation approach. 

Additional comments and edits:

Line 10 and 12 – spelling of acronym MAPaCMAn not MSPaCMAn

Line 92 – conical not conic

Line 108 – not applicable rather than inapplicable

Line 119-123 the statements around the use of the workflow don’t make sense as they are currently written.

Line 157 – the use of the word primordial.  This is not the correct word to use here.

Line 168-127 – agreed, you should do some statistical analysis, others have done work using grain size analysis from X-ray CT see the references listed below

https://doi.org/10.1016/j.mineng.2015.03.026

https://doi.org/10.1016/j.mineng.2015.06.001

line 191 – rather than writing “here it is shown” refer to the figure

Line 255 – spelling of parisite

Line 428 – spelling of Buzwagi

Line 450 – awkward phrasing, suggest starting the sentence as “In contrast to…

Line 470 – associated with each other

Line 593 – suggest using the term constituent phases rather than composing phases

Line 655 – awkward phrasing, suggest “classify its constituent phases, which makes it possible to distinguish…

Other comments:

How are the outputs from the analysis being validated?

I note that narrow size classes have been selected for the three samples, why have these been selected?

What is the rationale for the material and size ranges used for the spacer material?

The workflow in presented is not particularly novel – what is of interest are the sample preparation methods used and the particle by particle classification, which are only dealt with superficially, but it seems that these will be the subject of future papers? 

Author Response

We appreciate the time dedicated to the review of our article. We have largely followed your suggestions and added additional information to the methods section to meet the reviewer’s concerns.

This has the potential to be a very interesting piece of work, but at the moment it feels like there are missing parts to the paper.  The abstract and introduction suggest a methods paper (in particular sample preparation and particle characterisation) but there is limited detail on the rationale for the preparation methods chosen and validation of the particle characterisation approach. 

Reply: We added additional information to section 2.3 to link the validation method with the type of sample and also about the rationale for the spacer size and type to section 2.4.

We realise that this paper is not presented in the traditional way CT methods are usually presented. Typically, papers are focused on a method designed to improve a particular aspect of CT analysis using one specific material that is analysed under optimized conditions, and finally the results can be validated by some analytical technique. Instead, our paper does not focus on a specific aspect of CT analysis but rather on a workflow (or sequence of steps that can be optimized individually) that is meant to be general and not specific to the material. Therefore, the point of the paper is not to provide optimized methods for each of the sample types but rather to demonstrate the benefits and versatility of a new approach to analyse 3 different materials. Therefore, validation in this context is only qualitative.

Additional comments and edits:

Line 10 and 12 – spelling of acronym MAPaCMAn not MSPaCMAn

Reply: Corrected

Line 92 – conical not conic

Reply: Corrected

Line 108 – not applicable rather than inapplicable

Reply: The sentence was changed also as suggested by reviewer 1.

Line 119-123 the statements around the use of the workflow don’t make sense as they are currently written.

Reply: We hope the improved description of the workflow and it’s division into steps makes more sense now.

Line 157 – the use of the word primordial.  This is not the correct word to use here.

Reply: Changed to important.

Line 168-127 – agreed, you should do some statistical analysis, others have done work using grain size analysis from X-ray CT see the references listed below

https://doi.org/10.1016/j.mineng.2015.03.026

https://doi.org/10.1016/j.mineng.2015.06.001

Reply: Grain size is not something that is described by our method because the individual grains inside a particle are not segmented in the image (only the particle). Therefore, geometrical properties can only be measured at the particle level and phase compositions from all grains inside a particle without discrimination about how the volume of the phase is distributed, e.g. as one grain or multiple grains. We could speculate that grain segmentation would also be facilitated by our method, but that is not the focus of the paper.

line 191 – rather than writing “here it is shown” refer to the figure

Reply: Changed

Line 255 – spelling of parisite

Reply: The word was corrected all over the manuscript.

Line 428 – spelling of Buzwagi

Reply: Corrected

Line 450 – awkward phrasing, suggest starting the sentence as “In contrast to…

Reply: Done

Line 470 – associated with each other

Reply: Corrected

Line 593 – suggest using the term constituent phases rather than composing phases

Reply: Done

Line 655 – awkward phrasing, suggest “classify its constituent phases, which makes it possible to distinguish…

Reply: Corrected, that sounds much better, thanks.

Other comments:

How are the outputs from the analysis being validated?

Reply: This information was add to section 2.3. Samples 1 and 3, the classification was validated by MLA. Sample 2, the classification was compared to the theoretical attenuation coefficient of the known phases in the sample.

I note that narrow size classes have been selected for the three samples, why have these been selected?

Reply: No especial reason. We think it is interesting to show several examples to demonstrate the versatility of the method and to discuss some expected differences in the particle histogram due to the particle size (e.g. the broader peaks). Again we note that the purpose of this study is not the optimization of the workflow to any of the samples. The specific purpose of this work is to demonstrate the advantages and versatility of the method to classify phases inside particles.

What is the rationale for the material and size ranges used for the spacer material?

Reply: Additional information was add to the methods section to justify the spacer materials and size fractions used. “Sugar PMMA and graphite with different size fractions were used as spacer. Sugar has the advantages of being widely available, is cheap and can be easily crushed and sieved into specific size fractions. However, it is soluble in water and organic solvents used for polishing, thus it is not appropriate to be used is the sample is to be correlated to MLA. PMMA and graphite are insoluble in water, thus are used when the sample is used for MLA. The disadvantage of PMMA is that it is difficult to crush and to sieve into specific size fractions, and the graphite is difficult to obtain pure in larger size fractions.” And “i.e. size fractions smaller than the sample’s particle sizes help creating space between particles and size fractions about the same size of the sample’s particles help preventing the gravity segregation of denser particles.”

Additional discussion was also add to section 4.1 about how to possibly optimize the spacer particle size for the sheelite ore and the Qz/Py sample for which deviations from the ideal sample preparation are observed.

The workflow in presented is not particularly novel – what is of interest are the sample preparation methods used and the particle by particle classification, which are only dealt with superficially, but it seems that these will be the subject of future papers? 

Reply: We hope our changes to the paragraph associated to Figure 1 have clarified what we mean by workflow. It should be clear now that the whole point of the sample preparation is to ease particle segmentation that can then be analysed individually (as a workflow or a sequence of steps applied sequentially). We recognize that the individual steps are possibly not novel (e.g. we cannot claim that segmenting individual particles is novel), but the sequential use of the different steps (as a workflow) is to our knowledge unique, and it benefits for phase classification were not yet explored in the literature.

Yes, certainly there will be follow-up papers that will focus more on the analysis of all particles, bulk quantification and statistical analysis as the sum of all particles in the sample.  

Reviewer 3 Report

Please, see attached document

Comments for author File: Comments.pdf

Author Response

We appreciate the time dedicated to the review of our article and your encouraging message. We have improved the manuscript by adhering to your three suggestions, although we have some remarks about the limitations of the method.

The authors present a manuscript focused on classifying mineral phases in 3D images of particulate materials obtained by x-ray computed tomography, jointly using the interrelation of the gray values of the voxels and the geometric and morphological properties of the particles.

The method consists of three stages: a) sample preparation, b) individual particle analysis, and c) phase classification. Optimized sample preparation is based on the premise of increasing particle dispersions to reduce imaging artifacts.

The subject is very interesting and novel. The background and the bibliographic review are complete. The implementation of the method and its application to three geological materials—that present particles with quite different mineralogical and geometric characteristics—are exposed in an orderly and detailed manner. The document can be read with fluency.

Only three important points should be addressed:

1. At different points in the manuscript (lines 119, 130, 188, 661), the term “mineralogical properties” is introduced, but it is confusing for the reader which properties the authors refer to. From the beginning of the manuscript, this aspect should be made clear.

Reply: Good point. It is now clarified that “The mineralogical properties consist of the type, amount and distribution of phases inside a particle.” In this paper however we focus on the “type”, which is the first step towards quantification of other mineralogical properties.

 

2. It would be essential to include a paragraph summarizing the limitations of the method. For example:

Reply: We have significantly expanded section 4.4 (renamed to Applications and further work) to explain the scope of application of the method, which addresses some limitations suggested by the reviewer.

1. a) Sample preparation involves grinding and sieving, and therefore the textural characteristics (which are very important in many geological studies) cannot be determined.

Reply: We do not see sample preparation as a limitation specific of our method because any analytical technique requires some kind of sample preparation. In fact our sample preparation is very flexible and can be adjusted to the particle size instead of having to adjust the particle size to the method’s requirements (as in most other techniques).

45. b) Specify in a more numerical fashion the difficulty of discriminating minerals within the same mineralogical group (for example, feldspars, micas, carbonates, sulfides), since this point is also fundamental for a large part of the work related to the characterization of geological materials. Based on the references provided by the authors themselves (for example, Barn et al. 2020), it is very complex to discriminate minerals that have differences of less than 6% in the attenuation coefficient using a beam of 45.5KeV of effective energy. Considering their results, it is important to know what the authors can predict concerning this limitation for minerals of the same mineralogic group (e.g., calcite, dolomite).

Reply: One of the advantages of our method is that in principle one could distinguish phases even if the attenuation coefficient was the same for as long as there is some measurable characteristic common to all particles carrying those phases, e.g. shape of the particle or phase association within a particle. On the other hand phases that could in theory be distinguishable just based on their attenuation coefficient, their grey-scale may in practice be very similar due to the particle geometry (e.g. quartz / feldspar), by the ore microstructure (e.g. millerite/jamborite) or even by imaging artefacts. Therefore, we do not feel comfortable to impose a limit value to the difference in attenuation coefficients as suggested by the reviewer.   

1. c) It is necessary to have preliminary data of the samples (for example, 2D images), which unfortunately makes analytical work more expensive and easily delayed.

Reply: This is also not a limitation of our method but in general of all CT methods. Actually, even 2D automated mineralogy requires preliminary input of mineral lists and chemical essays for mass balance to achieve accurate classification. It is also an advantage of our method that some of the preliminary information that may be available about a material can be used, which is not always possible using other classification methods. 

3. To facilitate reading, it is suggested to use the same colors to illustrate the same mineralogical groups (for example, carbonates: green color) in Figures 4 and 5.

Reply: We have changed the colours of figure 4 to match figure 5.

 

The authors are congratulated for the promising results obtained and the present manuscript.

Reply: We appreciate the encouragement.

Round 2

Reviewer 2 Report

No additional comments on revised manuscript